Accelerate Matlab calculation for multi-dimension matrix

2016 3 24
2 个回答
更新时间：2016 3 24
1 次查看（30 天）

0 个投票

I want to solve a problem in space (3-dimensional:x y and z coordinate) and time(t). In general, my problem is like the code below:
A=rand(220);
x=0:1:50;
y=0:1:50;
z=0:1:50;
t=0:1:50;
B=zeros(length(x),length(y),length(z),length(t));
for ix=1:length(x)
for iy=1:length(y)
for iz=1:length(z)
for it=1:length(t)
B(ix,iy,iz,it)=sum(sum(sum(sum((A.*A*x(ix))+((A.^3)*y(iy))+...
((A+2*A)*z(iz))+((A.^4)*t(it))))));
end
end
end
end
However, the looping proccess takes much time. I am afraid i can not use repmat to decrease the computational time since it will need large memory to store the variable. Do you have any suggestion to make the simulation faster?
Thank you very much.
Michael

请先登录，再回答此问题。

回答（2 个）

the cyclist
the cyclist 2016-3-24

0 个投票

A 50x50x50x50 array of doubles is only about 48 MB of storage. I see no memory problem in vectorizing the whole calculation.

1 个评论
显示 -1更早的评论 隐藏 -1更早的评论

Michael Lubis
Michael Lubis 2016-3-24
Yeah, you are right. However, i want to multiply matrix with size 220x220 (the matrix A in my command) for each space and time entity. If i use repmat for the vectorization, then i have to create a matrix with dimension 50x50x50x50x220x220. Then i need a lot of memory.
Furthermore, in the real problem, i want to do 3 hours simulation with time interval 0.5 second (t=0:0.5:3*3600) which actually will require more memory. There is also a posibility for me to increase the range for space with smaller interval (for example x=-50:0.5:50). Therefore, is there another way to do all of this calculation instead of using looping?

请先登录，再进行评论。

Roger Stafford
Roger Stafford 2016-3-24

0 个投票

As you have written your code, A is a 2D array and yet you have taken four successive sums, which makes no sense. Either you meant A to be 4D or you meant to have only two successive sums. Either way your code is enormously inefficient. The summation of A.*A, A.^3, A+2*A, and A.^4 can be done just once for each of these four rather than 48,400 separate times each.
I will assume that A is to be 2D and you meant only two sums.
A=rand(220);
x=0:1:50;
y=0:1:50;
z=0:1:50;
t=0:1:50;
A1 = sum(sum(A.^2))*x;
A2 = sum(sum(A.^3))*y;
A3 = sum(sum(3*A))*z;
A4 = sum(sum(A.^4))*t;
B = zeros(51,51,51,51);
for ix = 1:51, for iy = 1:51, for iz = 1:51, for it = 1:51
B(ix,iy,iz,it) = A1(ix)+A2(iy)+A3(iz)+A4(it);
end, end, end, end
This should be much faster than your original code (unless Mathworks' compilers are awfully smart.)

1 个评论
显示 -1更早的评论 隐藏 -1更早的评论

Michael Lubis
Michael Lubis 2016-3-24
Yeah you are right. It should be only two sums. And exactly, your scheme calculate faster, eventhough still take takes sometimes for it to finish. Thank you very much for your answer.

请先登录，再进行评论。

类别

在 帮助中心 和 File Exchange 中查找有关 Resizing and Reshaping Matrices 的更多信息

产品

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by